[Mark Warren (ZeSnarf)]

Description

parseClaudeSessionFile reads a transcript with readFileSafe, which returns null for any file over MAX_FILE_SIZE (50 MB). Those sessions were left out completely, and they tend to be the longest and most important ones. Most of their size is raw api-request records, which the parser never reads.

This PR:

• streams the transcript with forEachJsonlLine instead of reading it whole;
• skips api-request lines before parsing them as JSON;
• applies MAX_FILE_SIZE to the lines it keeps. Past that limit it stops with a warning and parses what it has, instead of dropping the session;
• bumps CACHE_VERSION, so sessions that were skipped before are picked up.

We found this by running the extension on real Claude Code logs: the two biggest sessions, over 50 MB each, were missing from every view.

Related Issues

Fixes #364.
None filed.

Checklist

• npm run check passes (typecheck + lint + spellcheck + knip + tests)
• Changes are covered by tests (parser-claude.test.ts: a transcript over the limit, made of API request records, still parses its turn)
• Documentation updated (if applicable): not needed

[Mangesh Raut (mangeshraut712)]

CLAUDE_API_REQUEST_LINE = /^\s*\{\s*"type"\s*:\s*"api-request/; strictly anchors to the start of the line, requiring "type" to be the very first key immediately following {.

If a transcript's serializer places any other property first (e.g. {"timestamp": ..., "type": "api-request", ...} or {"uuid": ..., "type": "api-request", ...}), CLAUDE_API_REQUEST_LINE.test(rawLine) returns false.

Because the size accumulation happens prior to JSON parsing:

keptBytes += rawLine.length;
if (keptBytes > MAX_FILE_SIZE) {
  truncated = true;
  return;
}

any large API record with reordered keys will fail the regex filter, have its full multi-megabyte payload added to keptBytes, and prematurely trigger truncated = true (aborting further streaming via shouldStop).

Consider removing the opening brace anchor (e.g. /"type"\s*:\s*"api-request/ or checking rawLine.includes('"api-request"')) so API request records are bypassed before size accounting regardless of key order.

[Mark Warren (ZeSnarf)]

Thanks Mangesh Raut (@mangeshraut712), good spot. I've changed it to look for "type": "api-request" anywhere in the line, so the key order doesn't matter anymore. Added a test for that too, plus one that makes sure a prompt that just mentions it still gets parsed, since quotes inside strings are escaped and can't match.